BenChaliah/Superposition-Transformer
a novel architecture that leverages Autoencoders to superimpose the hidden representations of a base model and a fine-tuned model within a shared parameter space. Using B-spline-based blending coefficients and autoencoders that adaptively reconstruct the original hidden states based on the input data distribution.
Language:Jupyter Notebook
Total stars: 261
Stars trend:
1 Aug 2025
3am ▏ +1
4am ▍ +3
5am ▋ +5
6am █▍ +11
7am ▊ +6
8am ▉ +7
9am ▋ +5
10am █▏ +9
11am █▍ +11
12pm █▏ +9
1pm █▏ +9
2pm ▊ +6
#jupyternotebook
#artificialintelligence, #autoencoders, #deeplearning, #languagemodel, #llm, #llmtraining